Optical amplifiers produce spectrally related variations in gain that can be reduced by gain flattening filters having compensating spectral responses. The gain flattening filters, which are sometimes required to exhibit complex response profiles, can be assembled from filter components that exhibit simpler response profiles.
The complex spectral responses that are sometimes required of gain flattening filters can be approximated by concatenating filter components with simpler filter responses. Such compound gain flattening filters can be formed by concatenating conventional filters with spectral response profiles distinguished by their peak losses and their central wavelengths.
Best fit algorithms solve for the peak loss and central wavelength characteristics of the filter components to approximate a target loss spectrum. However, manufacturing even simple filter components with Gaussian response profiles can be difficult to achieve to desired accuracy. The resulting filter errors (i.e., insertion loss error function), which are largely due to unintended shifts in the central wavelengths of the filter components, can leave excessive gain ripple representing uneven amplification of the desired transmission spectrum.
Long period gratings with Gaussian response profiles can be used as filter components of compound gain flattening filters. Manufacturing these filter components to central wavelength accuracies less than one-half nanometer is difficult, and errors of only one Angstrom can produce significantly deteriorated results. Discarding gratings outside of tolerance and trial and error tuning are two approaches to meeting desired tolerances. Both are time consuming and expensivexe2x80x94the first due mainly to wasted efforts and the second due mainly to repeated efforts.
We have discovered that compound gain flattening filters can be assembled from a series of simpler filter components having wider tolerances than those normally prescribed but without the expected errors. Residual gain ripple from new amplifier and filter combinations can be minimized by approaching a best fit loss spectrum for the filters. Higher manufacturing yields are possible by utilizing filter components that would normally be considered outside acceptable tolerance.
Our new compound gain flattening filters can be manufactured using certain conventional steps including determining a filter desired loss spectrum and using a conventional algorithm to fit the desired loss spectrum with a series of filter components having individual spectral responses with peak losses specified at different wavelengths. However, instead of manufacturing a single set of filter components matching these specifications, at least a first of the filter components is fashioned from a pair of filter sub-components with similar spectral responses. The peak losses of the filter sub-components sum to the intended peak loss of the first filter component, and the central wavelengths of the filter sub-components also average to the specified central wavelength of the first filter component. However, the central wavelengths of the filter sub-components are offset in opposite directions with respect to the specified central wavelength of the first filter component for utilizing filter sub-components with wider range tolerances. The resulting response of the compound gain flattening filter is closer to the desired loss spectrum than if the central wavelength of the first filter were similarly offset in either direction.
Compound gain flattening filters can be made in accordance with our invention by dividing an entire series of filter components into two sets of filter sub-components. The central wavelengths of all of the filter sub-components of the first set are shifted in a positive direction and the central wavelengths of the filter sub-components of the second set are shifted in a negative direction, both with respect to the prescribed central wavelengths of the filter component series. Together, the two sets of filter sub-components produce an actual spectral filter response that is closer to the desired loss spectrum than if the central wavelengths of the series of filter components were similarly shifted in either direction.
The two sets of filter sub-components are particularly useful for reducing gain ripple of paired amplifiers having similar gain outputs, especially amplifiers separated by a fiber span. One set is positioned within one of the amplifiers, and the other set is positioned within the other amplifier. The gain ripples of the two amplifiers are separately reduced, avoiding amplifier-to-amplifier transmissions that are not appropriately corrected.
The filter sub-components of our invention are preferably made according to usual manufacturing practices, but are sorted following manufacture according to their shift from prescribed central wavelengths. For example, those shifted in a positive direction can be paired with those shifted by a similar amount in a negative direction so that the average central wavelength of the pair approaches the prescribed central wavelength. A similar balance can be achieved by larger combinations of sub-components having central wavelengths distributed about a prescribed mean.
Instead of or in addition to sorting by pairs or other combinations, tuning can be used to intentionally offset the central wavelength of one member of a combination to balance an opposite direction of offset exhibited by one or more other members of the combination. Tuning can also be practiced upon the one member to balance the filter central wavelengths in a static or dynamic fashion to accommodate changing conditions of use.
The sensitivity of the actual filter response to central wavelength errors varies among the series filter components. Central wavelength errors of filter components with particularly sharp profiles or large losses can contribute to insertion loss errors more than other filter components with similar wavelength errors but with flatter profiles. Thus, wider tolerances of central wavelength can be accommodated for filter components with flatter profiles. In the practice of our invention, just the filter components having the sharper profiles can be subdivided into balanced wavelength sub-components for reducing the number of required filter components and sub-components without requiring stricter tolerances for the sharper profile filter components.
The desired loss spectrum for the compound gain flattening filters represents the best fit of a combination of theoretical filter components varying in peak loss and central wavelength. However, the best fit of the series of filter components is still an approximation of a target loss spectrum required to entirely eliminate gain ripple.